A corn plant core localization algorithm was developed based on high-fit oriented bounding boxes for complex natural environment corn canopy data, and a labeling strategy was proposed that can effectively reduce the lack of edge precision. To address the issues of insufficient labeling accuracy and weak multi-scale feature extraction in traditional object detection networks, a novel high-precision labeling strategy for YOLO v7OBB was proposed and an innovative learning convergent asymptotic feature pyramid network (LC-AFPN) was developed. Additionally, a color space filtering algorithm was used for canopy segmentation, and a gap-filling algorithm improved image quality. Spatial moments were utilized to accurately calculate the coordinates of the plant core, leading to the learning convergent asymptotic YOLO v7OBB network (LCA-YOLO v7OBB) for corn canopy targets detection. Validation on a complex corn field dataset revealed that LCA-YOLO v7OBB offered strong anti-interference capability and high plant core localization accuracy, with an average accuracy of 85.19% and precision and recall rates of 91.83% and 83.04%, respectively. Compared with 12 other rotating object detection networks, this model demonstrated the best overall performance. Moreover, validation on custom cucumber and eggplant datasets further confirmed its robust generalization ability. This model provided a theoretical basis and technical support for applications such as precision fertilization and agricultural machinery visual navigation.